AI Insight
This study developed a machine learning framework to automate the color grading of chrysoprase gemstones, addressing the longstanding problem of subjective and inconsistent color evaluation in the gemstone market. CIE L*a*b* colorimetric data were collected from 51 real chrysoprase samples and 676 synthetic green reference points, then used to train and compare several classification algorithms including logistic regression, neural networks, k-nearest neighbors, support vector machines, and random forest. Logistic regression was selected as the optimal model due to its interpretability and efficiency, achieving a macro F1-score of 99.59% in mixed cross-validation and correctly classifying all 51 real samples.
Why it matters
Automating gemstone color grading reduces reliance on subjective human judgment, which could improve consistency and transparency in gemstone valuation and trade. The framework is potentially adaptable to other gemstones, offering a scalable tool for the broader gemological and jewelry industry.
Understand the Science
by Yuansheng Jiang, Ying Guo, Vien Cheung, Pohsun Wang, Stephen Westland
The color of gemstones plays a pivotal role in determining their quality and significantly impacts their market value. However, inconsistencies in gemstone color evaluation, stemming from the subjective nature of color perception, have hindered standardization in the market. Chrysoprase, celebrated for its distinctive apple-green color, is no exception. To address these challenges, this study employs a machine learning-based approach to automate chrysoprase color grading. CIE (1976) L*a*b* data were measured for 51 chrysoprase samples and 676 green reference points generated using the GemDialogue Color Reference using an X-Rite SP62 spectrophotometer. K-means was applied for color clustering, with Fisher discriminant analysis used to validate the clustering results. Various machine learning algorithms, including logistic regression, neural networks, k-nearest neighbors, support vector machines, and random forest, were trained on labeled data to assign chrysoprase colors to different groups. Logistic regression and neural network achieved comparably high macro F1-scores, and logistic regression was ultimately selected due to its simplicity, interpretability, and computational efficiency. In independent evaluation on 51 real chrysoprase samples, all samples were correctly classified within the present dataset. Additional mixed cross-validation incorporating both synthetic and real samples yielded consistent performance (macro F1-score = 99.59%), further supporting the robustness of the proposed approach. These results demonstrate the feasibility of applying machine learning techniques to structured gemstone color grading. The proposed framework provides a reproducible approach for objective chrysoprase color evaluation and may be adaptable to other gemstones with comparable colorimetric characteristics, subject to further validation. A publicly accessible chrysoprase color grading application is available at: https://github.com/harden2009190006/Chrysoprasecolorclassifier.
Source: Chrysoprase color grading with machine learning: A systematic approach